Fix array_has_all and array_has_any with empty array

[LuQQiu]

Which issue does this PR close?

• Closes array_has_any(column, []) with empty array throws RowConverter column schema mismatch, expected Utf8 got Int64 #15038

What changes are included in this PR?

Fix array_has_any and array_has_all with empty array input

Are these changes tested?

Added the unit test in array.slt

Are there any user-facing changes?

NA

[LuQQiu]

@jayzhan211 Could you help approve the test workflows and review the PR? Thanks in advance

[westonpace]

It's weird to me that All is true and Any is false but I agree it matches the descriptions:

array_has_all: Returns true if all elements of sub-array exist in array.
array_has_any: Returns true if any elements exist in both arrays.

I guess, from a set perspective, we have "true if the intersection of the two sets has the same length as needle" (all) and "true if the intersection of the two sets is non-empty" (any).

[jayzhan211]

D select array_has_any([1,2,3], []);
┌────────────────────────────────────────────────────────────┐
│ array_has_any(main.list_value(1, 2, 3), main.list_value()) │
│                          boolean                           │
├────────────────────────────────────────────────────────────┤
│ false                                                      │
└────────────────────────────────────────────────────────────┘
D select array_has_all([1,2,3], []);
┌────────────────────────────────────────────────────────────┐
│ array_has_all(main.list_value(1, 2, 3), main.list_value()) │
│                          boolean                           │
├────────────────────────────────────────────────────────────┤
│ true                                                       │
└────────────────────────────────────────────────────────────┘

LGTM

[westonpace]

Sorry, yes. I just meant it was strange intuition (I was rambling). I was not arguing correctness. I agree this is good.

[westonpace]

Using BooleanBuffer::new_set and BooleanBuffer::new_unset should be slightly more efficient.

[LuQQiu]

Using

    if needle.values().len() == 0 {
        return match comparison_type {
            ComparisonType::All => Ok(Arc::new(BooleanBuffer::new_set(haystack.len()))),
            ComparisonType::Any =>  Ok(Arc::new(BooleanBuffer::new_unset(haystack.len())))
        };
    }

Runs into

the trait bound `BooleanBuffer: arrow::array::Array` is not satisfied
the following other types implement trait `arrow::array::Array`:
  &T
  BooleanArray
  DictionaryArray<T>
  FixedSizeBinaryArray
  FixedSizeListArray
  GenericByteArray<T>
  GenericByteViewArray<T>
  GenericListArray<OffsetSize>
and 10 others
required for the cast from `std::sync::Arc<BooleanBuffer>` to `std::sync::Arc<dyn arrow::array::Array>`rustc[Click for full compiler diagnostic](rust-analyzer-diagnostics-view:/diagnostic%20message%20[0]?0#file:///Users/lu/Projects/Github/Work/datafusion/datafusion/functions-nested/src/array_has.rs)

So confusing about the different types loll

[alan910127]

You should use BooleanArray::from(BooleanBuffer::new_/* ... */)

[LuQQiu]

Thanks for the suggestion! Changed to use BooleanBuffer